Walter,
Thank you for this quote. This is exactly what Vladimir and I were trying to
say all along
in recent discussions on this topic... See, for example, slide 8 in
http://www.ibis.org/summits/feb18/dmitriev-zdorov.pdf
or the last slide in:
http://ibis.org/macromodel_wip/archive/20180227/arpadmuranyi/Referencing%20Problems/ReferencingProblems_2018_02_27.pdf
to mention a couple of slides. I am glad to hear that the "Who's Who of signal
integrity
and s-parameter experts" have the same views on this topic.
Our problem is that we have a few "complications" related to this in BIRD189
which
is what makes it hard for us to write clean and simple rules in for IBIS.
One is that our shorthand syntax for S-parameter model instantiation is defined
with
a single, common reference terminal (N+1) for all ports of the S-parameter
model.
Given the requirement that all ports which are connected together have to use
the
same reference (regardless of what that reference actually is), this implies
that all
N+1 terminal S-parameter models which are connected together will end up having
to use the same reference connection. This requirement can only be broken when
an S-parameter block appears with more than a single reference terminal, acting
as
an "isolation transformer" between the "sides". In our case, this can happen
if the
model maker uses of an IBIS-ISS subcircuit in which the S-parameter model is
instantiated with independent reference terminals, or if the EDA tool does
similar
S-parameter instantiations between the various IBIS components on the board
level.
We were struggling with defining rules in a few simple words for the different
situations, i.e. when the content of the IBIS-ISS subcircuit allows for
multi-referenced
S-parameter blocks, while the shorthand S-parameter syntax is based on the
single
reference concept. The rules will all of the sudden have a lot of IF-s and
BUT-s
which is hard to write down in a simple way. Or, if we over simplify the
rules, for
example saying that all references should be connected to node 0, we would be
making things too restrictive.
Add to this the situation with the BIRD158 buffer model schematic, which has a
"hard coded" reference terminal to that infamous (now) global reference node
(triangle symbol). Since the S-parameter block in BIRD158 uses a common
reference (N+1), anything that is connected to that will also have to use that
same global reference node. (Not because of the question whether there is
any current flowing through that reference node, but because ports connected
to each other must have the same reference).
So our rules would go something like this:
IF there is a BIRD158 buffer model then...
IF the on-die interconnect or package model(s) use the N+1 shorthand, then...
IF the on-die interconnect or package model(s) use IBIS-ISS subcircuits, but
only
IF those subcircuits contain S-parameter instances with more than one
reference, then...
IF the IBIS-ISS subcircuit contains other types of T-line models or discrete
elements, then...
IF the model is intended to be used for power aware simulations, then...
and the list might go on for a while...
So our problem is not the question whether we connect the reference terminals
of the S-parameter blocks to a global or a local reference node, it is not
whether
there is any current flowing through that reference connection, it is not
whether we
require the entire system to be referenced to a single reference node (the 3rd
drawing
in BIRD158), etc... Our problem is that we can't write a few simple and
parseable rules
for all of the possible situations which may arise with the various buffer and
interconnect
modeling options we support. And, if I understand it correctly, the latest
version of
BIRD189.5 does not guide the model maker with any rules on this, allowing the
inexperienced model maker to write models with gross errors.
I am not saying that we must have such rules in the spec, after all, one can
write all
kinds of ridiculously bad models in SPICE too, but I wonder if such rules would
be
useful to at least guard against the worst of such bad possibilities.
Thanks,
Arpad
====================================================================
From: ibis-interconn-bounce@xxxxxxxxxxxxx
[mailto:ibis-interconn-bounce@xxxxxxxxxxxxx] On Behalf Of Walter Katz
Sent: Friday, April 27, 2018 11:36 AM
To: 'IBIS-Interconnect' <ibis-interconn@xxxxxxxxxxxxx>; IBIS-ATM
<ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-interconn] What P370 say about S-Parameter referencing
All,
The following comes from the currently unapproved draft of:
P370(tm)/D2
Draft Standard for Electrical Characterization of High-Frequency Interconnects
at Bandwidths up to 50 GHZ
I point out the following lines from the statement below:
1. Current flowing in the signal terminal is equal to current flowing out of
the reference terminal at each port.
2. "the local references can be different for different ports of a
multiport, but they have to coincide with the local reference terminals of
another multiport to be connected."
The key point is that if two s-parameters are connected, then each s-parameter
measurement (or extraction from a layout data base) must use the same reference
at each port that is connected. This is a fundamental part of using s-parameter
data. It could be on a "ground" net or a power rail net, just needs to be the
same. Since they are connected, and the flow in and out of both of the
s-parameters are the same, then there is no net current flowing from the
connected s-parameters. So for simulation purposes, the actual "reference node"
connected to the S element is academic. I see no reason to challenge what the
P370 committee is stating, since the committee members are a Who's Who of
signal integrity and s-parameter experts.
[https://lh4.googleusercontent.com/ElUqNClc9FT7o1jGla7mphEBJd4g9WoUufRVtMmQEVvxxUF9N9cJrLCBw65P_J4-7R9dB72vBF_D-cJoD2GQgqgci7GC9pcDxss5y8fwX1Vxec7rVgGVjLvVKidxH_Z2FsvizjhmIZCBD_w8iw]
Figure B.1 Multiport currents and voltages definition
Ports are numbered from 1 to N. Each port has two terminals. One terminal is
the signal and another terminal is local reference or local ground terminal.
Current flowing in the signal terminal is equal to current flowing out of the
reference terminal at each port. Connection of multiports requires that the
definition of the terminals on both sides of the connection is the same. It
means that the local references can be different for different ports of a
multiport, but they have to coincide with the local reference terminals of
another multiport to be connected. Some groups of ports may share the reference
terminals. The currents and voltages at ports are either actual measurable
values or effective voltages and currents as in microwave theory. In frequency
domain, the currents and voltages are complex variables and can be united into
vectors with N complex elements:
Walter
Walter Katz
wkatz@xxxxxxxxxx<mailto:wkatz@xxxxxxxxxx>
978.461-0449 x 133
Mobile 303.335-6156